Apple revamps Siri after embarrassing mishaps, aiming to restore trust

Apple’s new “Campos” engine, a custom‑built version of Google’s Gemini large‑language model (LLM) with roughly 1.2 trillion parameters, will replace the rule‑based pipeline that has powered Siri since its 2011 debut. According to a February 25 report by John N., the shift to an LLM means Siri will move from matching spoken commands to a static list of intents toward generating responses in real time, a capability that underpins ChatGPT’s conversational fluency. Apple plans to run Campos on Google’s cloud infrastructure rather than on‑device, a decision that mirrors the deployment model used by most consumer‑facing LLMs today and reflects the massive compute requirements of a trillion‑parameter model.

The redesign is slated for iOS 27, which Mark Gurman of Bloomberg says will arrive this fall. Gurman notes that the updated Siri will be woven into core Apple apps, allowing users to ask natural‑language questions while browsing Photos, reviewing calendar events, or composing email. For developers, the change signals that voice‑driven interactions will soon be backed by a context‑aware AI that can parse on‑screen references and background conversation—a capability demonstrated in a recent Apple research paper highlighted by VentureBeat, where the company’s AI system “understands ambiguous references to on‑screen entities as well as conversational and background context.” That paper suggests the assistant will be able to disambiguate commands like “show me the one from last week” without requiring the user to specify a file name or location.

Apple’s pivot also marks a reversal of its long‑standing stance that users did not want a chat interface. The February 25 article points out that executives previously argued AI should operate silently in the background, but the competitive pressure from OpenAI’s hardware‑backed models and Google’s rapid Gemini releases forced a strategic rethink. Apple Intelligence, launched in 2024, “didn’t exactly blow anyone away,” the same source adds, with delayed features and half‑baked implementations that left analysts questioning whether the company had lost its edge. By adopting a proven LLM architecture and integrating it with its own on‑device perception stack—evidenced by the “see” AI research cited by VentureBeat—Apple hopes to close the gap with rivals such as Samsung, which already ships conversational AI baked directly into its phones.

From a technical standpoint, the move to a cloud‑hosted LLM raises questions about latency, privacy, and battery impact. John N. notes that while the model’s size enables richer understanding, the reliance on Google’s servers means Apple must negotiate data‑handling agreements that satisfy both regulatory scrutiny and user expectations for privacy. Apple’s existing on‑device neural‑engine expertise may mitigate some concerns, but the company has not disclosed whether any portion of the inference will be off‑loaded to the iPhone’s Apple Silicon chips—a detail that will likely surface in later developer briefings. The 9to5Mac piece on Apple’s speech‑generation research, which explores grouping similar sounds to speed up AI speech synthesis, hints that Apple is simultaneously optimizing the audio pipeline to keep response times snappy even when the heavy lifting occurs in the cloud.

Finally, the overhaul could reshape the mobile AI ecosystem. Developers will need to adapt to a Siri that can handle multi‑turn dialogs, generate images, and manipulate files through natural language, as Gurman predicts. Apple’s integration of Campos into its native apps suggests a tighter coupling between AI and the operating system, potentially reducing the reliance on third‑party SDKs for voice features. If the rollout lives up to the promise of “real, natural conversation,” Siri could finally shed its reputation as a party trick and re‑establish Apple as a serious contender in the consumer‑facing LLM race. The success of this effort will hinge on how well Apple balances cloud‑scale intelligence with the privacy and performance expectations that have long defined its brand.